mm: frontswap: remove unnecessary check during initialization
[linux-2.6-block.git] / mm / frontswap.c
CommitLineData
29f233cf
DM
1/*
2 * Frontswap frontend
3 *
4 * This code provides the generic "frontend" layer to call a matching
5 * "backend" driver implementation of frontswap. See
6 * Documentation/vm/frontswap.txt for more information.
7 *
8 * Copyright (C) 2009-2012 Oracle Corp. All rights reserved.
9 * Author: Dan Magenheimer
10 *
11 * This work is licensed under the terms of the GNU GPL, version 2.
12 */
13
14#include <linux/mm.h>
15#include <linux/mman.h>
16#include <linux/swap.h>
17#include <linux/swapops.h>
18#include <linux/proc_fs.h>
19#include <linux/security.h>
20#include <linux/capability.h>
21#include <linux/module.h>
22#include <linux/uaccess.h>
23#include <linux/debugfs.h>
24#include <linux/frontswap.h>
25#include <linux/swapfile.h>
26
27/*
28 * frontswap_ops is set by frontswap_register_ops to contain the pointers
29 * to the frontswap "backend" implementation functions.
30 */
31static struct frontswap_ops frontswap_ops __read_mostly;
32
33/*
34 * This global enablement flag reduces overhead on systems where frontswap_ops
35 * has not been registered, so is preferred to the slower alternative: a
36 * function call that checks a non-global.
37 */
38bool frontswap_enabled __read_mostly;
39EXPORT_SYMBOL(frontswap_enabled);
40
41/*
165c8aed 42 * If enabled, frontswap_store will return failure even on success. As
29f233cf
DM
43 * a result, the swap subsystem will always write the page to swap, in
44 * effect converting frontswap into a writethrough cache. In this mode,
45 * there is no direct reduction in swap writes, but a frontswap backend
46 * can unilaterally "reclaim" any pages in use with no data loss, thus
47 * providing increases control over maximum memory usage due to frontswap.
48 */
49static bool frontswap_writethrough_enabled __read_mostly;
50
51#ifdef CONFIG_DEBUG_FS
52/*
53 * Counters available via /sys/kernel/debug/frontswap (if debugfs is
54 * properly configured). These are for information only so are not protected
55 * against increment races.
56 */
165c8aed
KRW
57static u64 frontswap_loads;
58static u64 frontswap_succ_stores;
59static u64 frontswap_failed_stores;
29f233cf
DM
60static u64 frontswap_invalidates;
61
165c8aed
KRW
62static inline void inc_frontswap_loads(void) {
63 frontswap_loads++;
29f233cf 64}
165c8aed
KRW
65static inline void inc_frontswap_succ_stores(void) {
66 frontswap_succ_stores++;
29f233cf 67}
165c8aed
KRW
68static inline void inc_frontswap_failed_stores(void) {
69 frontswap_failed_stores++;
29f233cf
DM
70}
71static inline void inc_frontswap_invalidates(void) {
72 frontswap_invalidates++;
73}
74#else
165c8aed
KRW
75static inline void inc_frontswap_loads(void) { }
76static inline void inc_frontswap_succ_stores(void) { }
77static inline void inc_frontswap_failed_stores(void) { }
29f233cf
DM
78static inline void inc_frontswap_invalidates(void) { }
79#endif
80/*
81 * Register operations for frontswap, returning previous thus allowing
82 * detection of multiple backends and possible nesting.
83 */
84struct frontswap_ops frontswap_register_ops(struct frontswap_ops *ops)
85{
86 struct frontswap_ops old = frontswap_ops;
87
88 frontswap_ops = *ops;
89 frontswap_enabled = true;
90 return old;
91}
92EXPORT_SYMBOL(frontswap_register_ops);
93
94/*
95 * Enable/disable frontswap writethrough (see above).
96 */
97void frontswap_writethrough(bool enable)
98{
99 frontswap_writethrough_enabled = enable;
100}
101EXPORT_SYMBOL(frontswap_writethrough);
102
103/*
104 * Called when a swap device is swapon'd.
105 */
106void __frontswap_init(unsigned type)
107{
108 struct swap_info_struct *sis = swap_info[type];
109
110 BUG_ON(sis == NULL);
111 if (sis->frontswap_map == NULL)
112 return;
f9f08103 113 frontswap_ops.init(type);
29f233cf
DM
114}
115EXPORT_SYMBOL(__frontswap_init);
116
117/*
165c8aed 118 * "Store" data from a page to frontswap and associate it with the page's
29f233cf
DM
119 * swaptype and offset. Page must be locked and in the swap cache.
120 * If frontswap already contains a page with matching swaptype and
121 * offset, the frontswap implmentation may either overwrite the data and
122 * return success or invalidate the page from frontswap and return failure.
123 */
165c8aed 124int __frontswap_store(struct page *page)
29f233cf
DM
125{
126 int ret = -1, dup = 0;
127 swp_entry_t entry = { .val = page_private(page), };
128 int type = swp_type(entry);
129 struct swap_info_struct *sis = swap_info[type];
130 pgoff_t offset = swp_offset(entry);
131
132 BUG_ON(!PageLocked(page));
133 BUG_ON(sis == NULL);
134 if (frontswap_test(sis, offset))
135 dup = 1;
ef383597 136 ret = frontswap_ops.store(type, offset, page);
29f233cf
DM
137 if (ret == 0) {
138 frontswap_set(sis, offset);
165c8aed 139 inc_frontswap_succ_stores();
29f233cf
DM
140 if (!dup)
141 atomic_inc(&sis->frontswap_pages);
d9674dda 142 } else {
29f233cf
DM
143 /*
144 failed dup always results in automatic invalidate of
145 the (older) page from frontswap
146 */
165c8aed 147 inc_frontswap_failed_stores();
d9674dda
SL
148 if (dup) {
149 frontswap_clear(sis, offset);
150 atomic_dec(&sis->frontswap_pages);
151 }
4bb3e31e 152 }
29f233cf
DM
153 if (frontswap_writethrough_enabled)
154 /* report failure so swap also writes to swap device */
155 ret = -1;
156 return ret;
157}
165c8aed 158EXPORT_SYMBOL(__frontswap_store);
29f233cf
DM
159
160/*
161 * "Get" data from frontswap associated with swaptype and offset that were
162 * specified when the data was put to frontswap and use it to fill the
163 * specified page with data. Page must be locked and in the swap cache.
164 */
165c8aed 165int __frontswap_load(struct page *page)
29f233cf
DM
166{
167 int ret = -1;
168 swp_entry_t entry = { .val = page_private(page), };
169 int type = swp_type(entry);
170 struct swap_info_struct *sis = swap_info[type];
171 pgoff_t offset = swp_offset(entry);
172
173 BUG_ON(!PageLocked(page));
174 BUG_ON(sis == NULL);
175 if (frontswap_test(sis, offset))
ef383597 176 ret = frontswap_ops.load(type, offset, page);
29f233cf 177 if (ret == 0)
165c8aed 178 inc_frontswap_loads();
29f233cf
DM
179 return ret;
180}
165c8aed 181EXPORT_SYMBOL(__frontswap_load);
29f233cf
DM
182
183/*
184 * Invalidate any data from frontswap associated with the specified swaptype
185 * and offset so that a subsequent "get" will fail.
186 */
187void __frontswap_invalidate_page(unsigned type, pgoff_t offset)
188{
189 struct swap_info_struct *sis = swap_info[type];
190
191 BUG_ON(sis == NULL);
192 if (frontswap_test(sis, offset)) {
ef383597 193 frontswap_ops.invalidate_page(type, offset);
29f233cf
DM
194 atomic_dec(&sis->frontswap_pages);
195 frontswap_clear(sis, offset);
196 inc_frontswap_invalidates();
197 }
198}
199EXPORT_SYMBOL(__frontswap_invalidate_page);
200
201/*
202 * Invalidate all data from frontswap associated with all offsets for the
203 * specified swaptype.
204 */
205void __frontswap_invalidate_area(unsigned type)
206{
207 struct swap_info_struct *sis = swap_info[type];
208
209 BUG_ON(sis == NULL);
210 if (sis->frontswap_map == NULL)
211 return;
ef383597 212 frontswap_ops.invalidate_area(type);
29f233cf
DM
213 atomic_set(&sis->frontswap_pages, 0);
214 memset(sis->frontswap_map, 0, sis->max / sizeof(long));
215}
216EXPORT_SYMBOL(__frontswap_invalidate_area);
217
96253444
SL
218static unsigned long __frontswap_curr_pages(void)
219{
220 int type;
221 unsigned long totalpages = 0;
222 struct swap_info_struct *si = NULL;
223
224 assert_spin_locked(&swap_lock);
225 for (type = swap_list.head; type >= 0; type = si->next) {
226 si = swap_info[type];
227 totalpages += atomic_read(&si->frontswap_pages);
228 }
229 return totalpages;
230}
231
f116695a
SL
232static int __frontswap_unuse_pages(unsigned long total, unsigned long *unused,
233 int *swapid)
234{
235 int ret = -EINVAL;
236 struct swap_info_struct *si = NULL;
237 int si_frontswap_pages;
238 unsigned long total_pages_to_unuse = total;
239 unsigned long pages = 0, pages_to_unuse = 0;
240 int type;
241
242 assert_spin_locked(&swap_lock);
243 for (type = swap_list.head; type >= 0; type = si->next) {
244 si = swap_info[type];
245 si_frontswap_pages = atomic_read(&si->frontswap_pages);
246 if (total_pages_to_unuse < si_frontswap_pages) {
247 pages = pages_to_unuse = total_pages_to_unuse;
248 } else {
249 pages = si_frontswap_pages;
250 pages_to_unuse = 0; /* unuse all */
251 }
252 /* ensure there is enough RAM to fetch pages from frontswap */
253 if (security_vm_enough_memory_mm(current->mm, pages)) {
254 ret = -ENOMEM;
255 continue;
256 }
257 vm_unacct_memory(pages);
258 *unused = pages_to_unuse;
259 *swapid = type;
260 ret = 0;
261 break;
262 }
263
264 return ret;
265}
266
69217b4c
SL
267static int __frontswap_shrink(unsigned long target_pages,
268 unsigned long *pages_to_unuse,
269 int *type)
270{
271 unsigned long total_pages = 0, total_pages_to_unuse;
272
273 assert_spin_locked(&swap_lock);
274
275 total_pages = __frontswap_curr_pages();
276 if (total_pages <= target_pages) {
277 /* Nothing to do */
278 *pages_to_unuse = 0;
279 return 0;
280 }
281 total_pages_to_unuse = total_pages - target_pages;
282 return __frontswap_unuse_pages(total_pages_to_unuse, pages_to_unuse, type);
283}
284
29f233cf
DM
285/*
286 * Frontswap, like a true swap device, may unnecessarily retain pages
287 * under certain circumstances; "shrink" frontswap is essentially a
288 * "partial swapoff" and works by calling try_to_unuse to attempt to
289 * unuse enough frontswap pages to attempt to -- subject to memory
290 * constraints -- reduce the number of pages in frontswap to the
291 * number given in the parameter target_pages.
292 */
293void frontswap_shrink(unsigned long target_pages)
294{
f116695a
SL
295 unsigned long pages_to_unuse = 0;
296 int type, ret;
29f233cf
DM
297
298 /*
299 * we don't want to hold swap_lock while doing a very
300 * lengthy try_to_unuse, but swap_list may change
301 * so restart scan from swap_list.head each time
302 */
303 spin_lock(&swap_lock);
69217b4c 304 ret = __frontswap_shrink(target_pages, &pages_to_unuse, &type);
29f233cf 305 spin_unlock(&swap_lock);
69217b4c
SL
306 if (ret == 0 && pages_to_unuse)
307 try_to_unuse(type, true, pages_to_unuse);
29f233cf
DM
308 return;
309}
310EXPORT_SYMBOL(frontswap_shrink);
311
312/*
313 * Count and return the number of frontswap pages across all
314 * swap devices. This is exported so that backend drivers can
315 * determine current usage without reading debugfs.
316 */
317unsigned long frontswap_curr_pages(void)
318{
29f233cf 319 unsigned long totalpages = 0;
29f233cf
DM
320
321 spin_lock(&swap_lock);
96253444 322 totalpages = __frontswap_curr_pages();
29f233cf 323 spin_unlock(&swap_lock);
96253444 324
29f233cf
DM
325 return totalpages;
326}
327EXPORT_SYMBOL(frontswap_curr_pages);
328
329static int __init init_frontswap(void)
330{
331#ifdef CONFIG_DEBUG_FS
332 struct dentry *root = debugfs_create_dir("frontswap", NULL);
333 if (root == NULL)
334 return -ENXIO;
165c8aed
KRW
335 debugfs_create_u64("loads", S_IRUGO, root, &frontswap_loads);
336 debugfs_create_u64("succ_stores", S_IRUGO, root, &frontswap_succ_stores);
337 debugfs_create_u64("failed_stores", S_IRUGO, root,
338 &frontswap_failed_stores);
29f233cf
DM
339 debugfs_create_u64("invalidates", S_IRUGO,
340 root, &frontswap_invalidates);
341#endif
342 return 0;
343}
344
345module_init(init_frontswap);